A study of observer variability in CT Scan reading and coding was implemented utilizing identical sets of CT Scan polaroids and the CT Scan Data Collection form of the Pilot Traumatic Coma Data Bank. Readers were neurosurgeons participating in the Coma Data Bank. The degree of agreement among readers was calculated by Kappa statistics and an item analysis was performed. Severity of errors was determined by a substantive analysis. The application of Kappa statistics to this type of study is widespread, yet the restrictive assumptions of this method are rarely met. An extension of the usual methods, to allow for a fixed, not random, set of raters, is being developed.